How to implement that is written in the doc. I can't find code reference.

[miyannishar]

how to implement these evaluations types in the tools trajectory.

[llalitkumarrr]

Hello @miyannishar

Could you please elaborate on your question a bit more so that we can understand your query clearly and assist you better?

[miyannishar]

I am implementing the Evaluation of my Agentic System using pytest and I want to test the agent if its calling the tools in a series. Here in the docs its mentioned that we can evaluate the agent's tool call using exact match, in order, any order precision and so on so forth but for implementation, we only have one criteria: tool_trajectory_avg_score that only look for exact match. My question is how to implement the evaluation of tool call based on any order match instead of exact match only.

I am using LiteLLM, so using LLMAsAJudge is not an option for me.

[llalitkumarrr]

Hello @miyannishar,

Here on this page, you can find all the Evaluation Criteria available with ADK. If the tool_trajectory_avg_score metric doesn’t meet your requirement, you could explore rubric_based_tool_use_quality_v1. It is important to note that this evaluation method uses an LLM as a judge.

At present, these two are the recommended criteria for evaluating tool trajectories.